Start with ownership, not the bill
Before looking for savings, map every subscription, resource group and significant resource to an owner and a purpose. Unowned resources are both a cost problem and an operational risk: nobody is watching them, and nobody knows whether they can be turned off.
Separate production from everything else
Production and non-production environments deserve different controls, budgets and schedules. Many cost findings come from non-production resources that run around the clock at production sizes. Many reliability findings come from production resources that were configured like a test environment.
Review utilization against real requirements
Rightsizing is only useful when it is tied to recovery objectives, performance needs and growth plans. Document what each workload actually requires before changing tiers, so that a cost reduction does not quietly become an availability problem.
Make deployment and recovery part of the review
How changes reach production, and how you get back when something goes wrong, are part of operating cost. Check that deployments are repeatable, that backups are restorable, and that someone has practiced the recovery steps.
Prioritize by impact and effort, not by a savings target
A useful review ends with a ranked list of findings, each with evidence, a suggested owner and an effort estimate. Promising a percentage reduction before examining the environment sets the wrong incentive; a clear, prioritized plan is the deliverable.